A Review of Software for Predicting Gene Function - SERSC

International Journal of Bio-Science and Bio-Technology Vol.7, No.2 (2015), pp.57-70 http://dx.doi.org/10.14257/ijbsbt.2015.7.2.06

A Review of Software for Predicting Gene Function Swee Kuan Loha, Swee Thing Lowa, Mohd Saberi Mohamada,*, Safaai Derisa, Shahreen Kasimb, Choon Yee Wena, Zuwairie Ibrahimc, Bambang Susilod, Yusuf Hendrawand, Agustin Krisna Wardanid a

Artificial Intelligence and Bioinformatics Research Group, Faculty of Computing, Universiti Teknologi Malaysia,81310, Skudai, Johor, Malaysia. b Faculty of Computer Science and Information Technology, Web Technology Department, Universiti Tun Hussein Onn, 86400, Parit Raja, Batu Pahat, Johor, Malaysia. c Faculty of Electrical and Electronics Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia. d Faculty of Agricultural Technology, University of Brawijaya, Jl. Veteran Malang, ZIP 65145, Indonesia. [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] * Corresponding author: Email address: [email protected], Tel: 60-7-5533153, Fax: 60-7-5565044.

Abstract A rich resource of information on functional genomics data can be applied to annotating the thousands of unknown gene functions that can be retrieved from most sequenced. High-throughput sequencing can lead to increased understanding of proteins and genes. We can infer networks of functional couplings from direct and indirect interactions. The development of gene function prediction is one of the major recent advances in the bioinformatics fields. These methods explore genomic context by major recent advances in the bioinformatics fields rather than by sequence alignment. This paper reviews software related to predicting gene function. Most of these programs are freely available online. The advantages and disadvantages of each program are stated clearly in order for the reader to understand them in a simple way. Web links to the software are provided as well. Keywords: Gene Function, Artificial Intelligence, Bioinformatics, Cancer, Functional Genomics

1. Introduction The increase in the amount of genomic sequence data thanks to the Genome Project has increased the numbers of unknown gene. Hence, it is necessary to develop computational approach to discover the functions of unknown genes using the vast amount of information that is produced by genomic projects, such as gene expression data. Gene expression is the process involving splicing that is the central dogma for the production of RNA and gene products [1]. The increase in functional proteomic and functional genomic methods has increased the need for analysis of gene function. Examples of functions that are focused on in this field are gene translation, transcription, interaction among proteins and others. Gene function is important in bringing information of RNA transcripts, function of DNA at the level of genes and protein products. Discovering gene function serves as a goal that reveals relationships between the

ISSN: 2233-7849 IJBSBT Copyright ⓒ 2015 SERSC

International Journal of Bio-Science and Bio-Technology Vol.7, No.2 (2015)

phenotype and genome of an organism. Hence, a blueprint about how each gene functions in the cellular and molecular process is clearly provided. This leads to the possibility of understanding which genes and how they are involved in a particular mutation that is implicated in a human genetic disease. The work of discovering gene functions is not only important to understanding the biology of a particular organism but also plays an important role in identifying potential therapeutic targets. Gene function is difficult to process in their natural languages. However, there are several research areas that provide assistance in understanding gene function, such as gene regulatory network construction. A gene regulatory network represents the functional interactions among genes in a graphical view and applies functional genomics to predicting novel gene functions [2]. Genetic interaction is very informative for interpreting gene-gene functional linkages, as stated in You et al. [3]. It is extensively exploited for the dissection of specific pathway structures and annotating gene functions. A readable compute program should be developed to deal with gene data that can automatically annotate the huge streams of sequence produced by the genome sequencing project, such as in Delcher et al. [4] and Alexandersson et al. [5]. According to Zhou et al. [6], one novel data mining algorithm is Ontology-based pattern identification (OPI). Gene function expression patterns are identified based on the best existing knowledge. Gene ontology is a widely used in the study of gene function prediction. There are three ontologies in gene ontology: molecular function, the cellular component and biological process. A number of programs have been built to search, browse and process Gene Ontology. The predicting gene function is one of the major challenges in the field of biology. Gene function prediction is essentially the study of a classification problem based on machine learning techniques. Computational prediction methods give assistance in suggesting a restricted set of candidate functions that are verified through experimentation. Moreover, new hypotheses are generated directly. Also, guidance in exploring promising hypotheses is one of the benefits of using computational methods in prediction. The great availability of biological data has led to new directions in predicting function from sequence information and genomic analysis. The prediction of gene function can be accomplished by integrating several data sources. Genes with similar functions may not be clustered as a group in the clustering process. The specific interaction between genes needs to be obtained. Hence, several programs have been developed which contribute to this task. In this paper, software used to predict gene function will be introduced and summarized. The popular gene function prediction programs, like GeneMark and GeneFAS will be introduced. Some web-based programs have their own databases to allow the user to insert a dataset, such as a protein query, to be analysed. Software is written in different languages such as C language and Perl language. The search for gene function is initiated either in protein form or web form. The advantages and disadvantages are stated in order to let the user understand in a simple and clear way. A link to a web page with the software is also provided. Which software is more reliable and performs better at predicting gene function will be explained. The user should concern him/herself about the dataset to be provided and the parameters to be used in the predictions. Several problems have been found in predicting gene function. Very large numbers of functional classes can be found in Gene Ontology (over 1000) and FunCat (over 100). Taxonomies of gene function included Gene Ontology (GO) and Functional Catalogue (FunCat). Gene Ontology is a functional term set with some annotated genes that is widely used in the research area. Gene Ontology is widely used to classify gene function [7]. Gene Ontology is comprised of biological process ontology, cellular component ontology and molecular function ontology. Each gene or protein is associated by Gene Ontology with some functional term. The output in Gene Ontology is constructed based on the direct acyclic graph structure. The predicting gene function for an unknown gene involves assigning some Gene Ontology functional term to it. In the machine learning

58

Copyright ⓒ 2015 SERSC


community, this classification is called a multi-label classification problem. In contrast, FunCat produces a hierarchical tree-like structured output. FunCat only concentrates on the functional process. Hence, it is more stable than Gene Ontology. This paper provides basic concepts about gene function and its importance in the field of research. This review paper provides and reviews several programs involved in predicting gene function. A brief description of each program is given in this review paper to allow users to gain better understanding of these programs. Web links that access these programs together with some other features such as the platform and the programming language implemented in the program are also provided in table form. Moreover, the main advantages and disadvantages of each program are listed and discussed in the discussion section.

2. Gene Function Prediction Tools Many programs have been proposed to infer gene function. Nowadays, gene function prediction software provides inference services based on many features such as sequence and structure information, functional ontologies and heterogeneous data sources. Table 1 shows the sources of the gene function prediction programs that are discussed in this paper. Table 1. Sources of Gene Function Prediction Software Gene function prediction software SNAPper Funcassociate

OntoBlast GAIN GeneFAS

GFINDer GOToolBox Blast2GO

Biopixie GeneMark

Phydbac “Gene Function Predictor” VIRGO

Author / Year

Web Link

Platform

Implemente d Language

Kolesov et al.[8] Berrizl et al. [9] Zehetner [10] Karaoz et al. [11] Joshi et al.[12]

http://pedant.gst.de/snapper

Webbased Linux, Windows

C

C

Masseroli et al.[13] Martin et al.[14] Conesa et al.[15]

http://www.medinfopoli.polimi.it/GFI NDer http://gin.univ-mrs.fr/GOToolBox

Myers et al.[16] Basemer and Borodovsky [17] Enault et al.[18]

http://neurolex.org/wiki/Category:Res ource:bioPIXIE http://exon.gatech.edu/

Webbased Window, Linux Window, Linux, MacOsX Webbased Webbased Window, Linux, MacOsX Linux, Windows MacOS X,Linux, AIX, Solaris Webbased

Massjouni et al.[19]

http://whipple.cs.vt.edu:8080/virgo

Webbased

Java


http://llama.mshri.on.ca/cgi/func1/fun cassociate http://functionalgenomics.de/ontogate http://genomics10.bu.edu/gain/supple ments/cv.html http://digbio.missouri.edu/genefas

http://www.blast2go.de

http://www.igs.cnrsmrs.fr/phydbac/indexPS.html

C, Perl

C++, Perl Java

Java Perl Java

PHP, Perl C

Perl

59


GOBlast SynFPS ChemGenome

HCGene

Prosecutor

GeneMANIA

GO-At PANTHER 7 PhyloProf PlasmoPredict

AraNet Eukaryotic GeneMark

Argot2 FunCoup

Wang and Mo[20] Li et al.[21] Singhal et al.[22] Valentini and CesaBianchi [23] Blom et al.[24] WardeFarley et al.[25] Bradford et al.[26] Mi et al.[27] Valentini et al.[28] Tedder et al.[29] Hwang et al. [30] Borodovsky and Lomsadze [31] Falda et al.[32] Alexeyenko et al.[33]

http://bioq.org/goblast

Linux

Perl

http://www.synteny.net http://www.scfbioiitd.res.in/chemgenome/download.ht m http://homes.di.unimi.it/~valenti/SW/ hcgene/node1.html

Windows Linux, Windows, Solaris Linux, Windows, MacOsX

C# C

http://www.prosecutor.nl

Window, Linux, MacOsX Webbased

Java

Webbased Webbased Webbased Webbased

Java

Webbased Sun Solaris, Unix, Linux, AIX Webbased Webbased

Java

http://www.genemania.org

http://www.bioinformatics.leeds.ac.uk /goat http://www.pantherdb.org http://dna.cs.byu.edu/phyloprof/cli.ph p http://www.bioinformatics.leeds.ac.uk /~bio5pmrt/PlasmoPredict/PlasmoPre dict.html http://www.functionalnet.org/aranet http://topaz.gatech.edu/eukhmm.cgi

http://www.medcomp.medicina.unipd .it/Argot2 http://funcoup.sbc.su.se/

R

Java

Java Java Java

C

Java Java

2.1. SNAPper SNAPper is software developed by Kolesov et al. [8] to predict gene function according to the conservation of gene order. A SNAPper search can be performed by inserting a protein sequence. SNAPper is part of PEDANT, a server analysis of the genome. Relational databases store PEDANT data and can be assessed by the SNAPper server if allowed. Properties of gene products can be corrected by using SQL queries. Parameters like number of gene neighbours and finding orthologs are set by default to decrease the number of false positives. SNAPper can be implemented in C or with a UNIX environment. The service is developed on the PEDANT, which is used as a genome analysis server. Each gene product can be subjected to searches for protein motifs, protein features and homology. 2.2. Funcassociate Funcassociate is a web-based program built by Berrizl et al. [9] to assist users in applying Gene Ontology (GO) attributes to characterize large datasets of genes. Its distinguishing features include a Monte Carlo simulation and the ability to handle ranked input lists. The source code was developed in C and Perl language. Currently, Funcassociate can support ten organisms (Shewanellaoneidensis, Vibrio cholerae,

60



Saccharomyces cerevisiae, Caenorhaebditiselegans, Arabidopsis thaliana, Schizosaccharomycespombe, Drosophila melanogaster, Rattusnorvegicus, Homo sapiens and Musmusculus). An adjusted p-value is estimated from 1000 simulated queries of the null hypothesis result. The ability to restrict the genes in the universe used in the analysis with other features is not found in other similar programs. It is written in C language, while Perl is used in the CGI script that deals with requests over the web. 2.3. OntoBlast Zehetner [10] introduced OntoBlast, which is an inner function of Ontologies TO GenomeMatrix software for functional prediction purposes. OntoBlast utilizes functional ontology and sequence similarities to predict gene function. The objective is to overcome the challenge from gene functional inference by integrating gene sequence data. Gene Ontology is used by OntoBlast to support the functional annotation process. Search frame and list frame are two major aspects of OntoBlast. For the search frame, a result from the BLAST server, which is based on sequence similarity, is generated. Meanwhile, a list of functional ontology terms with different weighting numbers is provided in the list frame. Then the BLAST result and ontology terms are associated together for further prediction of gene function. A smaller weighting number indicates strong evidence for the similarity of their sequence data. Two genes that share the functional annotation may indicate they are sharing a similar function. 2.4. GAIN Karaoz et al. [11] developed GAIN, the Gene Annotation using Integrated Networks program to predict gene function using a computational approach. It operates in a functional linkage network. In the graph, nodes represent genes and edges represent how they are connected. In GAIN, each gene is predicted independently and multiple predictions can be made for the same genes. A single organism of FLN is constructed by integrating functional genomic information. Several algorithms are included in GAIN to systematically propagate annotations. Most of them are semi-supervised learning machine languages. Positive and unknown examples are easy to generate in GAIN but it is hard to generate negative samples. Positive samples are gene-function pairs, to which unknown samples with some of the same categorization are considered.

2.5. GeneFAS GeneFAS was introduced by Joshi et al. [12] and stands for Gene Function Annotation System. It combines various types of high throughput biological data sources such as microarray data, interaction among proteins, information about protein complexes and functional annotation for gene function prediction based on a probabilistic method. GeneFAS assesses functional relationships in the Gene Ontology hierarchy and correlations between proteins from microarray data. GeneFAS carries out a two-step mechanism for functional predictions. The observed frequency related to the unknown functions of genes based on the high throughput data information is the a-priori probability. The estimations of a-priori probabilities are applied to the investigations of functional similarities and interactions between genes for each data set with high throughput. Subsequently, prediction of function for an uncharacterized gene is carried out based on the estimated a-priori probabilities.


61


2.6. GFINDer Masseroli et al. [13] developed Genome Function Integrated Discover (GFINDer), which is a web server for functional prediction. Annotations of functional categories are extracted by GFINDer from multiple data sources. The main feature of this software is providing an evaluation of the annotation relevance for estimating the functional bias that exists in a microarray gene set. The statistical significance values based on the gene’s membership in a particular functional category are calculated. This statistical analysis on extracted functional categories is then applied to classify the gene’s function. The three layer architecture of GFINDer includes a data layer, processing layer and user layer. Various types of data sources and annotations are handled by the data layer. The processing layer is the main layer responsible for managing client requests. However, the user layer provides a graphical user interface for users to access the processing layer. 2.7. GOToolBox Martin et al. [14] developed a collection of software and methods for gene function prediction called GOToolBox. PERL is the programming language used to write GOToolBox programs. Overrepresented and underrepresented gene terms in a dataset can be identified according to the Gene Ontology. The genomic frequency of Gene Ontology functional terms embedded with genes is computer for the creation of dataset in gene function prediction. GO-Family and GO-Proxy are two of the major functions in this toolbox. Those genes that are functionally related are clustered together by the GO-Proxy program using calculations of gene distance based on Gene Ontology annotation terms followed by a clustering algorithm. However, GO-Family is responsible for searching the genes that share the same Gene Ontology annotation terms with the gene in the user query based on measurements of functional similarity. 2.8. Blast2GO Conesa et al. [15] proposed a desktop application program written in the Java language for functional annotation, Blast2GO. It aims to provide high throughput and automated sequence annotation for unannotated sequences based on similarity searches and allows functions to be associated with annotated genes. Blast2GO finds homologs using BLAST software. Gene Ontology functional terms that are retrieved according to the Blast hit sequences are assigning to a query sequence based on an annotation rule. An annotation score which by possibility of abstraction and the highest similarity hit of Gene Ontology term is calculated for each Gene Ontology candidate term for annotation assignment. Gossip is used to introduce the gene functionality in Blast2GO. The main feature of this software is its ability to visualize the analysis of functional genes in graphic form. 2.9. Biopixie Myers et al. [16] developed bioPIXIE, which can predict the components in a novel network based on different types of input data but does not simply recapitulate known biology. A general probabilistic system is developed for query-based discovery of pathway-specific networks through integration of diverse genome-wide data. bioPIXIE is a biological network predictor for S. cerevisiae which is a comprehensive and public system for visualization, integration and analysis. It is publicly available over the World Wide Web. It is convenient for researchers to use. The advantage of this software is it can incorporate different types of genomic data. The result shows that bioPIXIE dramatically increases the number of network components recovered with any of the individual types of evidence allowed. Researchers are enabled to study the interactions and novel pathways of components.

62



2.10. GeneMark Basemer and Borodovsky [17] developed the GeneMark web software, which provides an interface to their program designed to predict genes in eukaryotic, prokaryotic and viral genomic sequences. The server can analyse approximately 200 prokaryotic and more than 10 eukaryotic using pre-computed gene models and species-specific versions of the software. Genes in prokaryotic sequences from novel genomes can be determined using models derived on the spot upon sequence submission. This can either be done using the full-fledged self-training program GeneMarkS or a simple heuristic approach. A database of re-annotations of more than 1000 viral genomes by the GeneMarkS program is also available from the website. In order to provide the latest versions of the gene models and software, the GeneMark website is frequently updated. 2.11. Phydbac “Gene Function Predictor” The Phydbac “Gene Function Predictor” is gene annotation software introduced by Enault et al. [18]. This gene function predictor links to the Phydbac web server, which is written in Perl language. It consists of two major operating modes which are prediction and database gathering. For prediction, a consensus profile for the query sequence is generated and profiles that have the greatest similarity are retrieved. In addition, the query sequence is compared with all organisms to identify conserved neighbours on chromosomes. For database gathering, results that are related to the processed organisms are gathered. The results of database gathering are then retrieved using gene names or annotation keywords. The level of prediction confidence is displayed discriminately in terms of colours and keywords. The gene ontology functional annotation assignment is carried out after the association partner is identified. 2.12. VIRGO Massjouni et al. [19] presented a function prediction web server called VIRtual Gene Ontology implemented in Java language, due to the massive existence of unknown gene functions. At first, microarray data are associated with molecular interaction data in order to generate a network of functional associations. Gene ontology annotations are embedded in each node in the network graph. Finally, the annotation labels are propagated across the interaction network to make predictions about gene function. This software employs Gene Annotation using Integrated Networks (GAIN), implemented in C++ language, as the prediction engine for functional genes. Leave-one-out cross validation is performed to evaluate the performance of the GAIN algorithm. Another significant feature of this service is confidence estimates are provided to indicate the quality or priority of prediction for each unknown gene. A propagation diagram is also provided through this prediction service. However, H. sapiens and S. cerevisiae are the two organisms that VIRGO currently supports, since very large datasets are available. 2.13. GOBlast Wang and Mo [20] presented a software system that combines Gene Ontology annotations and Blast search for gene functional inference called GOBlast. There are three Bioperl modules that together constitute GOBlast functionality. For the data input module, a nucleotide or amino acid sequence in FASTA format with Gene Ontology annotation terms is used as input. For the data processing module, the input sequence with gene ontology annotation is used to implement a Blast search for sequence alignment. The Gene Ontology terms within the Blast results are selected for the retrieval of Gene Ontology information in order to submit to the next module. For the data output module, the genes scoring the highest on homology with the query gene and associated annotation


63


terms are shown. The associated annotation terms provide a description of the specific function of the gene. 2.14. SynFPS Li et al. [21] developed SynFPS, the Synteny-based Function Prediction System, in C# programming language. According to resemblances in gene distribution, a set of weakly related genomes is grouped together using a K-means clustering algorithm. Genes with similar functions are identified using regular expressions rather than sequence similarity. Distances between genes are taken into account in identifying genomes that are closely related. Subsequently, negative and positive training datasets are extracted from each group for supervised learning on gene function by a support vector machine. The negative training dataset is the neighbours to positive data, whereas the positive training dataset consists of gene groups that were identified by the system previously when matching the genes that were added manually with regular expression. This program only aims for functional predictions in the whole genome context based on the synteny of genes or the order of gene conservation and distances between genes. 2.15. ChemGenome ChemGenome is ab-initio gene prediction software developed by Singhal et al. [22]. Physico-chemical properties are used to construct a three dimensional vector to predict genes in Prokaryotic Genomes by obtaining the stacking energies and base pairing for each codon from the report. There are 64 codons in the CG2 model, each of which is assigned to a biology interaction such as stacking, hydrogen bonding and propensity interaction based on molecular dynamic simulation. The parameters of the software are calculated from MD stimulation. ChemGenome shows better results in differentiating genes from non-genes at an equivalent level. It brings out the possibility of useful and unique ab initio. Its results show that differentiating genes from non-genes gain similar better than the knowledge model trained. ChemGenome can be run in a Linux environment. The input requires a FASTA format genome sequence. The threshold value for a small genome should be set at a lower value and vice versa. Several methods are used in ChemGenome which include swissprot space, protein space and DNA space. It will predict prokaryotic coding regions if the user enters all or part of a genome. 2.16. HCGene Valentini and Cesa-Bianchi [23] introduced Hierarchical Classification of Genes (HCGene) for gene function prediction. It is an R package software library for supervised gene function classification. Subtrees related to biological problems, subgraph extraction, gene labelling and multiple gene products are allowed in HCGene. It also supports collaboration with semi-supervised and unsupervised methods. HCGene can analyse ontology taxonomies, which are Functional Catalogue and Gene Ontology for labelling functional classes. It provides three major capabilities, namely, generation of multi-labels, graphing and data processing. HCGene associates functional class labels that are derived from ontology taxonomies and different types of biological data, such as gene expression data, to the queried genes for functional inference. Several methods can be included in the extraction of Gene Ontology subgraphs and Functional Catalogue subgraphs. 2.17. Prosecutor Prosecutor is a Java based gene function inference engine that is mainly used for prokaryotes and was proposed by Blom et al. [24]. It is parameter-free software. Prosecutor treats each functional category as individual; hence, a trusted set of protein interactions with function information is not necessary. It fully utilizes various annotation

64



data, gene expression data and extra biological knowledge in a genomic context to predict gene function. Prosecutor predicts the functions of genes based on an iterative guilt-byassociation principle without a fixed cutoff. Genes with unknown functions are linked to genes with annotation terms at a high association rate using a sensitive algorithm. The significance of all relationships between functional annotation and genes is measured. As a result, the predictions of gene function visualize as a force-directed network layout graph using a Prefuse toolkit. 2.18. GeneMANIA The GeneMANIA prediction server is a gene function prediction web interface introduced by Warde-Farley et al. [25]. GeneMANIA generates hypothetical gene function, gene list analysis and prioritization of genes. According to available genomic and proteomic information, genes that have similar function as the query gene are appended into the gene list and visualized as a functional linkage network. In addition, the value of a prediction can be represented by assigning it a weight. By default, the weighting method that is assigned based on the query genes is employed for weight assignment. This assigned weight of the query genes indicates the predictive value of each dataset. All non-query genes are scored using label propagation and finally, for ranking purposes. GeneMANIA currently supports six different types of organisms. 2.19. Go-At Bradford et al. [26] presented Gene Ontology prediction in Arabidopsis thaliana (GOAt), which is a two-step web-based application since Arabidopsis thaliana is still poorly characterized. It utilizes the confluence of five different data sources, namely, the gene neighbourhood, interaction information, phylogenetic profile, sequence information and co-expression data in order to make functional inferences. In the first step, the probability of a functional interaction with the gene queried by the user is applied to automatically construct a ranked gene list in descending order of the probability of an interaction with the query gene and an associated score of prediction for each function. The assignment of the prediction score is based on the observed frequency of a function in the gene list that most probably represents the query gene function. In the second step, the predicted gene function with the best score is chosen. 2.20. PANTHER 7 Mi et al. [27] proposed PANTHER 7, which denotes Protein Analysis Through Evolutionary Relationships, for functional inference. PANTHER 7 predicts gene functions according to relationships across evolution and genes with known functions determined from experiments. Complete genome sequences enable analysis of evolution and the understanding of phylogenetic relationships [34]. PANTHER 7 generates phylogenetic trees using multiple protein alignments and associates the trees with annotation terms to explain the evolution of gene function. The phylogenetic trees are estimated using multiple alignments of proteins by the MAFFT alignment program. Events of gene duplication and speciation can also be represented by enhanced phylogenetic trees. Gene duplication events and gene trees are predicted using the GIGA algorithm. Determination of orthologs is also included in the gene function predictions. 2.21. PhyloProf According to Valentini et al. [28], PhyloProf is a four-stage process which includes a fuzzy genome profile constructed for each participating genome, after which the fuzzy profile is discretized, initial profile data is de-noised and the intra-inter genome distance is calculated. A distance diagram is produced and divided into four areas: bottom right


65


quadrant, top left quadrant, main diagonal and bottom left corner. The most interesting genes are presented based on their intra-inter distances. The interface was built using a combination of CSS, XHTML and PHP server scripting. It performs well for small tasks like querying the database of nucleotides. PhyloProf places an emphasis on the user experience and an aesthetic user interface. The results page shows each query gene. The query gene indicates which genomes had a gene homolog. It is very simple to use and yet a very powerful program for gene regulatory networks. 2.22. PlasmoPredict Tedder et al. [29] presented a web-based gene function prediction program called PlasmoPredict, since the gene functions of Plasmodium falciparum are for the most part still unknown. This software utilizes various types of data in a genomic context, such as high throughput data and structure or sequence data for functional inference. The gene ID of Plasmodium falciparum acts as the input. The output is a gene list that is functionally related to the query gene. PlasmoPredict is implemented based on the guilt by association principle. Moreover, the transfer of gene ontology annotation is also provided in this software. The prediction results are visualized as a network graph by the Medusa program. Each network edge is assigned a confidence value that indicates how reliably the two genes share a similar function. The confidence value can be measured by Bayesian classifier and functional annotations. 2.23. AraNet Hwang et al. [30] developed AraNet, a network-based functional prediction software for the plant Arabidopsis thaliana operating through a web interface. AraNet provides the prioritisation of genes for the functional screening based on the candidate genes. Two search options are provided by the AraNet website which can either search for new candidates from the pathway or predict a gene function according to its neighbours in the network. AraNet produces a functional linkage network with probabilistic criteria via integrating heterogeneous data and modified Bayesian network. Each edge in the functional linkage network is a log-likelihood score to indicate the probability of the positive linkage representation. AraNet collects novel information regarding gene function from the network’s neighbours for prediction. 2.24. Eukaryotic GeneMark Borodovsky and Lomsadze [31] developed Eukaryotic GeneMark, one of several programs used to find eukaryotic genes. Two protocols are used to predict gene function in GeneMark program. The first protocol is called GeneMark.hmm. The second protocol is called GeneMark-ES, which performs self-training on the anonymous input sequence and an estimation of species specific parameters of the HMM model is generated. The programs GeneMark-ES and GeneMark.hmm-E are used to find genes in eukaryotic genomes. The GeneMark program may provide valuable information about analysis, especially in eukaryotic genomes. The latest version of the software can be run under the operating system of the LINUX environment. In comparing GeneMark-E and GeneMark.hmm-E, GeneMark-ES requires more resources. 2.25. Argot2 Argot2 is a program introduced by Falda et al. [32] to support gene functional inference at a large scale. It stands for Annotation Retrieval of Gene Ontology Terms. It provides a function prediction service through a web interface. The input of Argot2 is the list of sequences, which is in FASTA format. A weighting value is assigned to each annotation term based on its BLAST and HMMER e-values. Furthermore, the most

66



accurate gene ontology terms are selected by clustering methods and finally associated with those input sequences. Semantic similarities are taken into account in the processing of weighted gene ontology terms. The reference databases currently supported by Argot2 are Pfam and UniProt. The database server can be accessed by three different methods, i.e. interactive analysis, batch analysis and consensus analysis. 2.26. FunCoup FunCoup is a database developed by Alexeyenko et al. [33] that visualizes and maintains global protein or gene networks of functional coupling. Bayesian integration is used to construct the diverse high-throughput data. Functional coupling information is transferred between species. In version 2 of FunCoup, the new input datasets have been improved in term of network quality and coverage. There is a dramatic increase in the number of high-confidence network links. In the latest version, the human network has been increased eight-fold. FunCoup performs comparative interactomics such as aligning sub-networks between different species using orthologs. There are major increases in data of higher quality and in comprehensive data. This brings benefits such as an increase in total evidence and more accurate predictions. For example, using the FunCoup network, a gene’s functional relatedness to Alzheimer’s disease can be analysed or determined by the enrichment of common interactors.

3. Discussion Many programs have been designed for gene function prediction based on different criteria. However, each program has its own strengths and weaknesses. Some programs only support gene function prediction for one organism while others can support more than a single species. For example, AraNet only supports one organism, Arabidopsis thaliana. Then, GeneFAS only supports Sacchromycescerevisiae whereas the Phydbac “Gene Function Predictor” only supports Escherichia coli. In addition, there are only a very few programs that focus on plants. This may be due to the scattered resources and literature in the plant area [35]. Furthermore, some programs predict gene function based on a functional linkage network or phylogenetic trees. However, most programs use Gene Ontology annotation and integrate heterogeneous data sources to facilitate gene function prediction. Table 2 shows the main advantages and limitations of gene function prediction software. Table 2. Main advantages and limitations of gene function prediction software Gene function prediction software Funcassociate

Main advantages

Main limitations

The CGI script of FuncAssociate’s and is written in Perl while the software is written in C using standard modules

Funcassociate can support up to a maximum of ten organisms

OntoBlast

Allows comparison of all genes from nine organisms simultaneously

GAIN

Operating in functional linkage network so that multiple predictions can be made for the same genes Predictions can search by either partial or complete matching yeast gene names Provides evaluation of statistical significance of functional categories

Leads to false positive ontology terms due to insignificant sequence similarities Negative samples are hard to generate

GeneFAS GFINDer


Limited to S. cerevisiae The analysis and annotation are only as accurate as the

67


GOToolBox

Blast2GO Biopixie

GeneMark

Phydbac “Gene Fucntion Predictor” VIRGO

GOBlast

Provides GO-Diet software that allows a certain ontology depth to be selected so as to restrict GO terms and reduce the number of GO terms in forming a slim GO hierarchy Easy to distribute and low maintenance Multiple types of genomic data are integrated for 3 diverse KEGG pathways, and robust enhanced integration of different types of genomic data Pre-computed gene models can be processed and models derived on sequence submission identified prokaryotic gene sequences from novel genomes Useful for incomplete sequences due to blast mode Assigns confidence estimates to allow biologists to prioritize gene function predictions for further analysis Gene Ontology and Blast search are combined into a single step to save time

SynFPS

Determines genes that are functionally similar using regular expressions

ChemGenome

Only simple parameters are needed for predictions and the stability of hydrogen bonding between the DNA bases. Prokaryotic and eukaryotic genome gene finders are enhanced Capable of integrating heterogeneous data sources and processing graphs and multi-labels. Annotations for gene products and genes are given. Both hierarchies function at different degrees of resolution and multi classes involved are provided Parameter free and has additional prokaryote-specific information layers

HCGene

Prosecutor

SNAPper

GeneMANIA GO-At

PANTHER 7

68

SNAP function prediction is accepted for querying a protein sequence. Representation of hyperlinked graphical is rendered Provides gene prioritization and weight assignment Predicts unknown gene functions by determining ten putative golgins/Golgiassociated proteins A phylogenetic tree represents all the

employed annotation databases It does not support a directed acyclic graphical output option

Limited to Gene Ontology annotations Pre-processed data cannot be classified

Higher quality data are needed

Limited to Escherichia coli Only supports H. sapiens and S. cerevisiae currently The input sequences need to satisfy the format for the wublast program Only extendable to other organisms that have properties similar to phage genomes There is no different data is integrated

Does not have a hierarchical classification algorithm since it is more prone to facilitate gene classification algorithms

Only focus on prokaryotes and occurrence of false positives is proportional to the number of performed tests Definitive of function is not included in the result.

Only supports six different organisms currently The prediction performance for biological process and the cellular component still need improvement Only model domains that are



PhyloProf PlasmoPredict AraNet

Eukaryotic GeneMark

Argot2

FunCoup

evolutionary events in the gene family and experimental results keep changing A distance diagram is produced and divided into four areas Includes a wide range of functional genomics data Provides gene set enrichment analysis of the valid query genes

Gene prediction to find exact exon/intron and gene boundaries has improved using GeneMark.hmm-E and the absence of a training data problem has been solved using the GeneMark-ES algorithm Highly scalable and able to annotate from small gene sets to whole genomes Great increase in higher quality data; facilities for subnetwork conservation analysis are provided; possibility of performing comparative interactomics

conserved across a single subfamily Has not been well developed Only focuses on parasitology It is not suitable for some predictions of particular function and prediction power is influenced by genes that are not well connected A longer time cost for predictions

User needs to provide the BLAST and HMMER search results The need to set parameter values such as network distance and identify the gene set

4. Conclusion In the previous sections, numerous approaches for the computational prediction of gene function from various types of biomedical data were discussed. The analysis of biological data involves handling a number of challenges, many of which have only been partly addressed. Some of the most prominent challenges are widely varying sizes of functional classes and most classes being very small, hierarchical arrangement of functional labels as in Gene Ontology, and incompleteness and various types and extents of noise in biological data. Each type of biological data usually has a strong correspondence with a certain type of function that can be best predicted using data sets of that type. The Gene Ontology is increasingly being established as the most appropriate functional classification scheme for protein function prediction research because of its several desirable properties, and the forward looking attitude of its curators who are keeping it up-to-date with latest research. In particular, several GO-friendly approaches have recently been proposed, which incorporate the hierarchical structure of GO in the prediction technique so as to exploit the parent-child relationships between various functional classes. GO contains separate ontologies for three different types of protein function, namely cellular component, molecular function and biological process, thus making it easier to identify the most appropriate functional hierarchy to be used for making predictions from biological data of a certain type. Even though many advances have been made in the field of gene function prediction, there is still a lack of understanding of the most appropriate prediction technique for any particular category of genes. Thus, there is a great need for the creation of benchmark datasets and the adoption of a consistent evaluation methodology, standardization will help in the identification of the most appropriate function predictions strategy in a certain context, and the current weaknesses and needs of the field. Last but not the least, we firmly believe that an efficient scientific workflow can be established, in which, first, hypotheses are generated by executing the appropriate function prediction algorithm on the available biological data, then, these hypotheses are validated experimentally, thus leading to confident predictions of a gene’s function.


69


Acknowledgements We would like to thank the Universiti Teknologi Malaysia for supporting this research through a GUP research grant (Grant number: Q.J130000.2507.05H50) and a Matching grant (Grant number: Q.J130000.3007.00M27).

Conflict of Interest There is no conflict of interest in this manuscript.

References [1] [2] [3] [4] [5] [6] [7]

[8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]

[26] [27] [28] [29] [30] [31] [32] [33] [34] [35]

70

M. Bae, Y. Kim, J. Lee, Y. Jung and H. Kim, International Journal of Genomics, (2013). C. Li, M. Liakata and R. Schucmann, Briefings in Bioinformatics,vol. 15, no. 5, (2013). Z. H. You, Z. Yin, K. S. Han, D. S. Huang and X. B. Zhou, Bioinformatics, vol. 11, no. 343, (2010). A. L. Delcher, D. Harmon, S. Kasif, O. White and S. L. Saizberg, Nucl. Acids Res., vol. 27, no. 23, (1999). M. Alexandersson, S. Cawley and L. Pachter, Genome Research, vol. 13, no. 3, (2003). Y. Y. Zhou, J. A. Young, A. Santrosyan, K. S. Chen, S. F. Yan and E. A. Winzeler, Bioinformatics, vol. 21, no. 7, (2005). M. Ashburner, C. A. Ball, J. A. Blake, D. Botstein, H. Butler, J. M. Cherry, A. P. Davis, K. Dolinski, S. S. Dwight, J. T. Eppig, M. A. Harris, D. P. Hill, L. I. Tarver, A. Kasarskis, S. Lewis, J. C. Matese, J. E. Richardson, M. Ringwald, G. M. Rubinand and G. Sherlock, Nature Genetics, vol. 25, no. 1, (2000). G. Kolesov, H. Mewes and D. W. Frishman, Bioinformatics, vol. 18, no. 7, (2002). G. F. Berrizl, O. D. Kingl, B. Bryant, C. Sander and F. P. Roth, Bioinformatics, vol. 19, no. 18, (2003). G. Zehetner, Nucl. Acids Res., vol. 31, no. 13, (2003). U. Karaoz, T. M. Murali, S. Letovsky, Y. Zheng, C. Ding, C. R. Cantor and S. Kasif, Proceedings of the National Academy of Sciences of the United States of America, (2004). T. Joshi, Y. Chen, J. M. Becker, N. Alexandrov and D. Xu, OMICS, vol. 8, no. 4, (2004). M. Masseroli, D. Martucci and F. Pinciroli, Nucl. Acids Res., vol. 32, no. 2, (2004). D. Martin, C. Brun, E. Remy, P. Mouren, D. Thieffry and B. Jacq, Genome Biol., vol. 5, no. 101, (2004). A. Conesa, S. Götz, J. M. G. Gómez, J. Terol, M. Talón and M. Robles, Bioinformatics, vol. 21, no. 18, (2005). C. L. Myers, D. Robson, A. Wible, M. A. Hibbs, C. Chiriac, C. L. Theesfeld, K. Dolinski and G. Troyanskaya, Genome Biology, vol. 6, no. 13, (2005). J. Basemer and M. Borodovsky, Nucl. Acids Res., vol. 33, (2005). F. Enault, K. Suhre and J. Claverie, BMC Bioinformatics, vol. 6, (2005). N. Massjouni, C. G. Rivera and T. M. Murali, Nucl. Acids Res., vol. 34, no. 2, (2006). C. Wang and Z. Mo, Journal of Chongqing University, vol. 6, no. 3, (2007). J. Li, S. Halgamuge, C. Kells and S. L. Tang, BMC Bioinformatics, vol. 8, no. 6, (2007). P. Singhal, B. Jayaram, S. B. Dixit and D. L. Beveridg, “Biophysical Journal, vol. 94, no. 11, (2008). G.Valentini and N.C. Bianchi, Bioinformatics, vol. 24, no. 5, (2008). E. J. Blom, R. Breitling, K. J. Hofstede, J. B. T. M. Roerdink, S. A. F. T. V. Hijum and O. P. Kuipers, BMC Genomics, vol. 9, (2008). D.W. Farley, S. L. Donaldson, O. Comes, K. Zuberi, R. Badrawi, P. Chao, M. Franz, C. Grouios, F. Kazi, C. T. Lopes, A. Maitland, S. Mostafavi, J. Montojo, Q. Shao, G. Wright, G. D. Bader and Q. Morris, Nucl. Acids Res., vol. 38, (2010). J. R. Bradford, C. J. Needham, P. Tedder, M. A. Care, A. J. Bulpitt and D. R.Westhead, The Plant Journal, vol. 61, no. 4, (2010). H. Mi, Q. Dong, A. Muruganujan, P. Gaudet, S. Lewis and P. D. Thomas, Nucl. Acids Res., vol. 38, (2010). N.C. Bianchi and G. Valentini, Machine Learning in Systems Biology, vol. 8, (2010). P. M.Tedder, J. R. Bradford, G. A. McConkey, A. J. Bulpitt and D. R.Westhead, Trends in Parasitology, vol. 26, no. 3, (2010). S. Hwang, S. Y. Rhee, E. M. Marcotte and I. Lee, Nature Protocol, vol. 6, no. 9, (2011). M. Borodovsky and A.Lomsadze, Curr. Protoc.in Bioinformatics, (2011). M. Falda, S. Toppo, A. Pescarolo, E. Lavezzo, B. Di Camillo, A. Facchinetti, E. Cilia, R. Velasco and P. Fontana, BMC Bioinformatics, vol. 13, no. 14, (2012). A. Alexeyenko, T. Schmitt, T. Tharnberg, D. Guala, O. Frings and L. L.Sonnhammer, Nucl. Acids Res., vol. 40, (2012). H. C. Rawal, N. K. Singh and T. R. Sharma, International Journal of Genomics, (2013). V. Sharma and I. N. Sarkar,Briefings in Bioinformatics, vol. 14, no. 2, (2012).


A Review of Software for Predicting Gene Function - SERSC

A Review of Software for Predicting Gene Function - SERSC

Suggest Documents